Abstracts

Rationale: The origin of the broad bandwidth of frequencies comprising the human EEG spanning over four orders of magnitude (<0.05 Hz – ultraslow – to >500 Hz – fast ripples) remains controversial. In particular, there is uncertainty over the relative contribution of discrete oscillations at specific frequencies and neural ‘noise’ in generating the continuous EEG spectrum (Buzsaki, 2006; Rhythms of the brain). For epileptologists, the routes of transition from the broadband background EEG to the pattern of a seizure spectrum of discrete frequencies remains of fundamental importance. Here we formulate a unified statistical description of the awake and asleep electrocorticogram (ECoG). The model offers a phenomenological window into the network architecture of the cerebral cortex and its sleep-wake and normalcy-seizure transitions.Methods: Power spectra of paired wake and sleep ECoG epochs (sampled between 200-1000 Hz and low-pass filtered between 70-300 Hz) from five adult epilepsy patients undergoing invasive monitoring with subdural electrodes were computed. A five-component, three-parameter log-normal mixture model was fitted to the data (MATLAB Statistics Toolbox, Natick, MA). Statistical comparison between the model coefficients in the wake and sleep states was performed with the Welch test.Results: Uniformly good fits of ECoG spectra to the mixture model were obtained, with coefficient of determination (R2) values ranging from 0.5-0.9. The clearest difference between the coefficients of the wake versus sleep model fits was in the amplitudes of the individual component means, which was highly significant (p < 0.001). Minor, though also statistically significant, differences were observed between the location of the component means as well as their individual variances.Conclusions: (i) The proposed model offers statistical characterization of the entire ECoG spectrum, and in a manner that treats wakefulness and sleep along a continuum of parameters, and naturally includes the classical Berger bands. This approach contrasts with previous efforts to fit power-law distributions to bandlimited ECoG that ignore Berger bands (Milstein et. al., PLOS One 2009; 4(2): e4338). (ii) Wakefulness and sleep differ essentially in the distribution of power to the Berger bands, which otherwise vary little between the two states. (iii) The log-normal nature of the model generates the hypothesis of the background EEG arising fundamentally from an ensemble of ‘noisy’ oscillators that interact by multiplicative modulation, implying an algebraic interaction of their phases. (iv) Though speculative, this framework in turn suggests a mechanism for phase locking of brain rhythms underlying normal function as well as pathological (epileptic) hypersynchrony (Kalamangalam et. al., Clin Neurophysiol 2013; 125 (7): 1324-1328).